I have run the linpack benchmarks on my Linux system, and got
very good results.  I have never seen numbers this high for a 386-33:

Unrolled Double Precision Linpack

    times are reported for matrices of order   100
      dgefa      dgesl      total       kflops     unit      ratio
 times for array with leading dimension of  201
       1.49       0.04       1.53        449       4.46      27.32
       1.49       0.04       1.53        449       4.46      27.32
       1.49       0.05       1.54        446       4.49      27.50
       1.49       0.04       1.54        447       4.48      27.45
 times for array with leading dimension of 200
       1.52       0.05       1.57        437       4.57      28.04
       1.52       0.05       1.57        437       4.57      28.04
       1.52       0.05       1.57        437       4.57      28.04
       1.52       0.05       1.57        438       4.57      28.00
Unrolled Double  Precision   438 Kflops ; 10 Reps 
Unrolled Double Precision Linpack

     norm. resid      resid           machep         x[0]-1        x[n-1]-1
       1.7        7.41628980e-14  2.22044605e-16 -1.49880108e-14 -1.89848137e-14


My hardware is: 386DX-33MHz, with 64KB 0ws cache, and 8MB 4WS main
memory.  I have a cyrix 83D87-33Mhz coprocessor (new mask), ISA bus,
18ms IDE drive.

My software is Linux .98pl1, gcc 2.2.2.d (pl?), linpackc with -DDP and
-DUNROLL options.  I ran several times, and results varied by .003 or
so at most between runs.  These results came after a cold boot.

linpackc contains a separate subroutine at the end called second which
counts the execution time.  I had to change one line in this routine:
        from :"void getrusage();"
        to:   "extern int getrusage();"

Any comments?  I would mail them to Jack Dongarra, but I do not think
32 bit results are kept these days.

Robert Rosenbaum
rrosen@caip.rutgers.edu